Data

…will only work if you finished last set of notes (or open this week’s lab).

load("data/wrangled/wrangled_data_vaping.rda")

Question

When is pivoting your data from wide to long (or long to wide) helpful?

data |>
   pivot_longer(cols = columns_to_pivot , names_to = "new_col_for_labels" , values_to = "new_col_for_values")

EDA: Brainstorm

What do you want to know?

Notes from class:

• count:
  • by age groups - age distribution
  • Grade
  • Group
  • Sex: Males, Females, NA
• Relative proportion of e-cig use by gender (anecdote: spent cartridges in male br and not female in hs; hypothesis: male use higher than females, but want to see)
• Distribution of most popular brands (all brands)
• Distribution of most popuular flavors (all flavors)
• year on the x-axis; y is frequency of use; some measurement of use (e-cig, tobacco, …)

EDA: skim

library(skimr)
skim(nyts_data)

Note: If you include this in a report, you should also guide the viewer. Theres a lot in there. What do you want your reader to know?

Things discussed in class when looking at skim output:

• why brand_ecig has so many missing values (only have brand information for 2019)
• what mean calculation for a categorical value means (proportion of TRUE values)
• why Age min and max values look off (b/c it’s coded as a character due to >18 category)
• that the n category is how many respondents there are each year

Remember: It can be very helpful to thinkg “what overall trends do I see?” and “is there anything weird going on?” any time you’re looking at an EDA output

EDA: Basics

Getting a sense for some of the categorical data

table(nyts_data$Age)

  >18    10    11    12    13    14    15    16    17    18     9 
  727    50  5360 13499 14613 14036 13498 13205 12754  7108   198 

table(nyts_data$Group)

Combination of products                 Neither       Only e-cigarettes 
                  16517                   61738                    7866 
    Only other products 
                   9344 

table(nyts_data$year)

 2015  2016  2017  2018  2019 
17711 20675 17872 20189 19018 

Remember: If you include this in a report, you’ll also need text to explain what the reader should know/take away from any of these.

EDA: Plots

Sex by Year

nyts_data |>
  group_by(year) |>
  count(Sex) |> 
  ggplot(aes(x=year, y=n, color=Sex)) + 
  geom_col()

(Note: I argued that this was not the best way to display these data…but it’s a good start. If you include in your report, you likely want to improve!)

E-Cig use by gender

Student suggestion: Relative proportion of e-cig use by gender (anecdote: spent cartridges in male br and not female in hs; hypothesis: male use higher than females, but want to see)

• barplot between two genders; proportion of each (gender on x axis - proportion of use on y axis); doing the same across different tobacco use categories
• want to capture year in some way (facet)
• line plot (year on x-axis; measurement of use on y-axis (proportion), lines for Sex)

nyts_data |>
  group_by(year, Sex) |>
  # count things 
  summarize(mean_ever = mean(ecig_ever, na.rm=TRUE)) |>
  ggplot(aes(x=year, y=mean_ever, group=Sex, color=Sex)) +
  geom_line()

Brands

Student suggestion: Distribution of most popular brands (all brands)

nyts_data |> 
  group_by(year) |>
  count(brand_ecig)

# A tibble: 12 × 3
# Groups:   year [5]
    year brand_ecig     n
   <dbl> <chr>      <int>
 1  2015 <NA>       17711
 2  2016 <NA>       20675
 3  2017 <NA>       17872
 4  2018 <NA>       20189
 5  2019 Blu          111
 6  2019 JUUL        2028
 7  2019 Logic         36
 8  2019 MarkTen       32
 9  2019 NJOY          44
10  2019 Other       1253
11  2019 Vuse         100
12  2019 <NA>       15414

The above helps us remember that we only have brand information from 2019.

nyts_data |> 
  filter(year == 2019, !is.na(brand_ecig)) |> 
  ggplot(aes(x=brand_ecig)) +
  geom_bar()

Flavors

Student suggestion: Distribution of most popular flavors (all flavors)

This one is more complicated b/c flavor data are across multiple columns…going to leave this for the analysis set of notes

Tabacco use across time

Student suggestion: year on the x-axis; y is frequency of use; some measurement of use (e-cig, tobacco, …)

nyts_data |>
    group_by(year) |>
  # count things 
  summarize(mean_ecig_ever = mean(ecig_ever, na.rm=TRUE),
            mean_tobacco_ever = mean(tobacco_ever, na.rm=TRUE)) |>
  pivot_longer(-year, names_to = "variable", values_to = "values") |>
  ggplot(aes(x=year, y=values, group=variable, linetype=variable)) +
  geom_line()

Suggested Reading

• Case Study from OCS